基于视频序列的人体三维重建

3D Reconstruction of Human Body Based on Video Sequence

计算机视觉在当今信息社会下拥有众多的应用场景,被利用在很多领域中,如军事、安保和医疗等。其中三维重建的研究在计算机视觉的领域中占有极高的位置。本文立足于视频序列,提出了一种人体三维重建方法。该方法对若干段视频序列进行一系列的处理和加工,包括相机追踪、深度恢复等三维重建基本操作,最终可以得到一个人体三维模型。此外,我们提出了一种能够让各序列之间相似变换求解的有效程度得到保证的方法,即基于RANSAC的迭代分层式求解方法。针对纹理不丰富的区域深度恢复质量不高的情况,在深度恢复时,我们采用模版约束的方法来解决这个问题。本文还采用了一种新的方法用于模版对其过程,即通过人体模型的主成分分解来对齐主方向。结果表明,基于以上各种方法的人体三维重建方法可以获得较为理想的的人体模型。

Computer vision has numerous application scenarios in information society,which is used in many fields such as military,security and medical care.Among them,the study of 3D reconstruction occupies a very high position in the field of computer vision.Based on the video sequence,this paper proposes a method for 3D reconstruction of human body.The method performs a series of processing on several video sequences,including 3D reconstruction basic operations such as camera tracking and depth recovery,and finally a 3D human body model can be obtained.In addition,we propose a method that can guarantee the degree of validity of the similarity transformation between the sequences,that is,an iterative hierarchical solution method based on RANSAC.For the case that the texture depth recovery quality is not high in the region where the texture is not rich,we adopt the template constraint method to solve this problem in depth recovery.This article also uses a new method for the template alignment process,which is to align the main direction through the principal component decomposition of the human body model.The results show that the 3D reconstruction of human body method based on the above methods can obtain an ideal human body model.